Research On Vehicle Detection Technology And Application Based On Feature Point Trajectory Clustering

Traffic video surveillance system is a hot research topic in the field of intelligent transportation.At present,researchers have proposed many moving target detection algorithms.However,due to the complexity of traffic video scene,the detection results are easily affected by external factors such as camera shake,camera Angle change,vehicle occlusion,light change,etc.which brings great challenges to vehicle target detection based on traffic video.In view of the above situation,this paper proposes a vehicle detection method based on feature point trajectory clustering,which is as follows:(1)Feature point trajectory acquisition: for the traffic monitoring video,the ORB algorithm and pyramid Lucas Kanade optical flow algorithm are used to detect and track the moving target feature points,and the track set of moving target feature points is established.(2)Track 3D information extraction: combining with camera calibration and backprojection principle,the feature point track is backprojected onto a plane in the 3D world coordinate system to estimate the 3D information of the track.Based on the analysis of the rigid relation of the trajectory of vehicle feature points,the inverse projection velocity,relative height and trajectory position are used to construct the feature vector.(3)Trajectory initial clustering and inter-class merging based on spectral clustering: according to the height constraint between vehicle trajectories,the similarity matrix between trajectories is sparsely processed,and the trajectory of feature points is preliminarily clustered by Ncut criterion.Based on the three-dimensional model of vehicle,the results of spectral clustering were combined among classes to obtain the final vehicle detection results and vehicle trajectories.(4)Traffic information extraction: by analyzing the three-dimensional information in the vehicle trajectory,the judgment of the individual behavior of the vehicle and the extraction of traffic flow parameters of the section can be realized.The above research scheme effectively solves the problem of vehicle detection and traffic parameter acquisition in traffic video surveillance system.The detection process combined with the 3D information of the trajectory of feature points effectively improves the real-time detection accuracy of vehicles to 93%.The traffic parameters obtained by analyzing the movement trajectory are of great application value for in-depth analysis of the traffic condition of road sections and improvement of the safety and capacity of road traffic.